Download Design and Clinical Evaluation of a Handheld Wavefront Autorefractor

1040-5488/15/9212-1140/0 VOL. 92, NO. 12, PP. 1140Y1147
OPTOMETRY AND VISION SCIENCE
Copyright * 2015 American Academy of Optometry
ORIGINAL ARTICLE
Design and Clinical Evaluation of a Handheld
Wavefront Autorefractor
Nicholas J. Durr*, Shivang R. Dave*, Fuensanta A. Vera-Diaz†, Daryl Lim*, Carlos Dorronsoro*,
Susana Marcos*, Frank Thorn†, and Eduardo Lage*
ABSTRACT
Purpose. To introduce a novel autorefractor design that is intended to be manufacturable at low cost and evaluate its
performance in measuring refractive errors.
Methods. We developed a handheld, open-view autorefractor (the ‘‘QuickSee’’ [QS]) that uses a simplified approach to
wavefront sensing that forgoes moving parts and expensive components. Adult subjects (n = 41) were recruited to undergo
noncycloplegic refraction with three methods: (1) a QS prototype, (2) a Grand Seiko WR-5100K (GS) autorefractor, and (3)
subjective refraction (SR). Agreements between the QS and GS were evaluated using a Bland-Altman analysis. The accuracy
of both autorefractors was evaluated using SR as the clinical gold standard.
Results. The spherical equivalent powers measured from both autorefractors correlate well with SR, with identical correlation coefficients of r = 0.97. Both autorefractors also agree well with each other, with a spherical equivalent power 95%
confidence interval of T0.84 diopters (D). The difference between the accuracy of each objective device is not statistically
significant for any component of the power vector (p = 0.55, 0.41, and 0.18, for M, J0, and J45, respectively). The spherical
and cylindrical powers measured by the GS agree within 0.25 D of the SR in 49 and 82% of the eyes, respectively,
whereas the spherical and cylindrical powers measured by the QS agree within 0.25 D of the SR in 74 and 87% of the
eyes, respectively.
Conclusions. The prototype autorefractor exhibits equivalent performance to the GS autorefractor in matching power
vectors measured by SR.
(Optom Vis Sci 2015;92:1140Y1147)
Key Words: refraction, wavefront aberrometry, autorefractor, refractive errors, global health
U
ncorrected refractive errors are a leading cause of disability globally and are especially prevalent in low-resource
settings.1 One of the major barriers to obtaining prescription
eyeglasses in these settings is that eye care is often inaccessibleVthere
is a shortage of eye care professionals and the efficiency of existing
providers is limited by a lack of equipment.2 Autorefractors are used
ubiquitously in high-resource settings to expedite the refraction process and enable minimally trained personnel to measure eyeglass
prescriptions, but they are too expensive for widespread adoption in
low-resource settings.
*PhD
†
OD, PhD
Madrid-MIT M+Visión Consortium, Massachusetts Institute of Technology,
Cambridge, Massachusetts (NJD, SRD, DL, EL); Department of Biomedical
Engineering, Johns Hopkins University, Baltimore, Maryland (NJD); New England
College of Optometry, Boston, Massachusetts (FAV-D, FT); Instituto de Óptica
‘‘Daza de Valdés,’’ Consejo Superior de Investigaciones Cientı́ficas, Madrid, Spain
(CD, SM); and Instituto de Investigaciones Biomédicas ‘‘Alberto Sols,’’ UAM/CSIC,
Madrid, Spain (EL).
An eyeglass prescription contains a minimum of three numbers
that characterize the refractive power of each eye (the ‘‘refraction’’):
the spherical power, S, cylinder power, C, and the axis, A. The six
parameters (three for each eye) obtained with an autorefractor are
often used as a starting point for subjective refraction (SR) and
to reduce the overall time required of an examiner to obtain an
eyeglass prescription. Standard autorefractors use high-quality
optical components with moving parts and are traditionally large,
stationary systems. Portable autorefractors such as the Nikon
Retinomax are also available, but these are more expensive and less
accurate than standard tabletop systems.3,4 There are also emerging
low-cost techniques for objective refraction that use adjustable
lenses2,5,6 or a modified mobile phone,7 but these approaches have
yet to demonstrate comparable accuracy to commercial autorefractors or the gold standard of SR.
This study introduces and clinically evaluates a new, customdeveloped, handheld autorefractor called the QuickSee (QS).
It is based on Shack-Hartmann wavefront sensing, but unlike
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Handheld Wavefront AutorefractorVDurr et al.
conventional wavefront aberrometers, the QS is composed of
low-cost components, a simplified optical layout, and no moving
parts. The prototype is also open view, which reduces the influence of instrument-induced myopia. Refractions obtained from
the QS prototype were compared with those obtained with a
Grand Seiko WR-5100K (GS) open-view autorefractor and with
binocular SR. The GS autorefractor model chosen for comparison
in this study is frequently used in other clinical studies and known
to agree well with SR.8,9
METHODS
QS Prototype
The QS prototype tested in this study is a new, simplified
implementation of a Shack-Hartmann wavefront aberrometer.
It differs from conventional wavefront aberrometers in a number
of ways: (1) it uses a laser diode as a light source instead of a superluminescent diode, (2) it uses a low-cost complementary metal-oxidesemiconductor detector instead of a scientific-grade charge-coupled
device for the wavefront sensor, (3) there is no Badal relay lens system
in either the illumination or detection paths, (4) there is no dedicated
pupil camera, and (5) it is a see-through device. These modifications
allow for a simplified layout (Fig. 1A); a compact, handheld form
factor (Fig. 1B); and a reduced bill of materials. Much of the reduction in performance that would result from the removal and
reduction in quality of components compared with a conventional
wavefront aberrometer is mitigated with the use of a custom data processing algorithm that examines many recorded wavefront images to
calculate the refractive power of the eye.
The removal of the Badal system in our aberrometer does create
three complications that have the effect of reducing the measurement range, but not significantly reducing performance within the
measurement range. First, the illumination beam is always collimated entering the pupil, resulting in large spots on the retina for
eyes with severe primary aberrations. Because the retinal plane is
approximately conjugate with the image sensor plane, this can result
in large spots in the Shack-Hartmann image that are difficult to
1141
isolate. Second, without moving lenses to compensate for defocus
aberration, the envelope of the wavefront may converge or diverge
as it enters the microlens array (Fig. 1A, inset). Thus, for severe
myopia, the wavefront will be undersampled by the lenslet array, and
for severe hyperopia, the periphery of the pupil is not sampled.
Third, because the pupil plane is not optically conjugate with the
microlens array in our configuration, the refraction calculated at the
microlens array must be propagated to the spectacle plane using a
vertex correction.
The QS prototype uses an 850-nm laser diode that delivers an
average power to the eye of 250 KW with a beam diameter, Db,
of 1 mm at the cornea. The safety threshold for viewing 850-nm
collimated light for 5 minutes continuously is 332 KW, as
specified by the ANSI Z136.1-2000 Standard for protection of
the human eye.10 The laser diode creates a point on the retina of
the subject, the light from which is then remitted through the pupil,
passed through an 850-nm band-pass filter with a 10-nm full width
at half maximum, and analyzed with a custom wavefront sensor. The
choice of a relatively high power source enabled a high signalto-noise ratio of the spots in the Shack-Hartmann image at fast
imaging rates. This proved to be especially important in patients
with strong refractive errors where the beacon generated on the
retina is blurred. Additionally, a bright source facilitated alignment
of the prototype for the subject. We also include a 1.0 neutral density
filter at the output port of the prototype that attenuates the ambient
light in the see-through channel to increase the contrast of the source
for the subject.
The propagation distance from the pupil to the lenslet array, ll,
is 105 mm. The wavefront sensor is composed of a 1.2-mm-thick,
10- 10-mm microlens array and a 1.3-megapixel monochrome
complementary metal-oxide-semiconductor image sensor with a
5.3- 6.7-mm active area. Each lens in the microlens array is
plano on one side and convex parabolic on the other side with a
focal length, fl, of 19 mm and a pitch, p, of 300 Km. The QS
prototype connects to an external power source and a laptop
computer that records and processes wavefront images.
Eye refractions are calculated automatically, with no human
intervention, using a custom algorithm that analyzes the sequence
FIGURE 1.
(A) Layout of the QS low-cost open-view autorefractor. An 850-nm laser diode illuminates a point on the retina (dashed red line) and a custom-built
wavefront sensor captures the remitted light field (solid blue lines). This schematic shows the wavefront envelope simulated with Zemax from a 4-mmdiameter pupil of an emmetropic eye. The inset shows a magnified view of the wavefront envelope from eyes that are j4 D myopic, +4 D hyperopic,
and emmetropic. The width of the image sensor (6.66 mm) shows the size of the wavefront envelope relative to the size of the sensor. (B) Photographs of the
QS in use show that the layout can be implemented in a compact prototype that can be held and aligned by the subject. A color version of this figure is
available online at www.optvissci.com.
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
1142 Handheld Wavefront AutorefractorVDurr et al.
of wavefront images recorded by the QS. To optimize the exposure of the recorded wavefront images, a closed-loop control
system adjusts the frame rate (from 5 to 25 Hz) and exposure time
(from 40 to 200 ms) of the detector to maintain a target percentage (0.1%) of saturated pixels in the image. After the video
acquisition is complete, the algorithm removes unusable frames
from the raw image sequence using a set of quality metrics. The
most common causes for unusable frames in the 41 patients tested
here were (1) misalignments between the eye and the device, (2)
low signal owing to saturated pixels from corneal reflections, and
(3) no signal attributed to the subject blinking. Frames are also
removed if they contain less than 21 resolvable spots, which is the
minimum required for fitting a sixth-order Zernike polynomial.
An average of 281 T 113 frames was recorded per video. After
removing unusable frames, there was an average of 137 T 93
wavefront images remaining to analyze in each video. The input
data to the QS algorithm are thus substantially different from conventional wavefront aberrometers, which typically calculate aberrations from a single frame rather than from a dynamic sequence of
wavefront images.
Wavefront images that pass the filtering stage are evaluated
individually to create a set of refractions and then the statistics
of the set are analyzed to generate the final prescription. Depending on the number of spots identified in the image, the slopes
of the wavefront sampled within each lenslet are fitted to a 6thto 10th-order Zernike polynomial expansion using least-squares
estimation.11 The second-order Zernike coefficients are used
to determine the power vector representation of the refraction
(M, Jackson cross-cylinder components at 0 degrees, J0, and at
45 degrees, J45),12 using well-known formulas.13 The magnitudes
of some higher-order coefficients are known to correlate with
visual acuity,14 but we found that evaluating only the low-order
aberrations resulted in accurate autorefraction. The software
groups the refractions in sets with similar M, J0, and J45 (binned at
0.125-diopter [D] increments), and the average value in the data
set with the largest number of elements (the statistical mode of
the histogram) is used as the uncalibrated refraction value.
The QS is calibrated by measuring refractions from an emmetropic model eye with known trial lenses placed at the spectacle
plane to vary M, J0, and J45, independently. The model eye has a
5-mm-diameter iris in front of a 19-mm-focal-length doublet lens
(Thorlabs AC127-019-B-ML) with a flat surface placed approximately at a focal distance behind the lens to act as the retina. Values
from these uncalibrated measurements are interpolated piecewise
and applied to each power vector component measured by the QS
to generate the calibrated output. In addition to this calibration, an
additional offset is expected to account for the difference in dispersion between the model eye and human eye. The model eye has
a defocus shift of only 0.25 D (calculated in Zemax) whereas a
human eye is expected to have a defocus shift of about 0.75 D
between 850 and 543 nm.15 However, in our pilot human tests, we
found that we did not need to include an offset to account for the
0.5-D difference in the QS refractions to agree with the GS and SR.
This discrepancy is attributed to the model eye being emmetropic
at green wavelengths, which was later confirmed by a measurement
of j0.5 D of myopia by a commercial autorefractor.
The myopic range of the QS is limited by ray crossoverVwhen
the wavefront converges so strongly that spots formed on the image
sensor from adjacent lenslets cannot be resolved. For a system like
the QS, with no compensation optics between the eye and the lenslet
array or the light source, Campbell16 has used geometric optics to
derive a relation for the minimum refractive error measurable, Pmin:
1
Cs
Db Il e
P min ¼ j
j
= 1þ
ð1Þ
2f l 2p
2pIl p In
where Cs is the tangent of the angle subtended by the source, le is
the distance between the retina and the second principal plane of
the eye, lp is the distance between the retina and the second nodal
point of the eye, and n is the index of refraction of the eye. Using
values of Cs = 0.0067, le = 23 mm, lp = 17 mm, and n = 1.336, the
theoretical Pmin of the QS prototype is j5.7 D. This agrees well
with the measured limit of j6 D determined by measuring trial
lenses with a model eye.
In hyperopic eyes, the wavefront envelope expands as it propagates from the pupil to the lenslet array. Thus, the hyperopic
range of the QS, Pmax, is limited by the diameter of the pupil, Dp,
desired to be sampled and the diameter of the wavefront sensor,
Dl. The maximum measureable refractive power can then be determined by geometric optics to be:
P max ¼
Dl jDp
:
Dp Il l
ð2Þ
The smallest dimension measured by the QS wavefront sensor
is 5.3 mm. In this model, the QS can measure up to 7 D of hyperopia
while still sampling at least a 3-mm diameter of the pupil.
Experimental Protocol
Institutional review boards at both the New England College of
Optometry and Massachusetts Institute of Technology approved
the protocol used in this study. The study also conformed to the
tenets of the Declaration of Helsinki. Subjects were recruited from
the students and staff at the New England College of Optometry via e-mail requests to approximately equally populate four
refractive groups categorized by their spherical equivalent power,
M: high myopes (j6.00 D e M G j3.00 D), low myopes (j3.00
D e M G j0.50 D), emmetropes (j0.50 D e M G 1.00 D), and
hyperopes (1.00 D e M G 4.00 D) as measured with SR. Additional criteria for inclusion were (1) no history of surgery or eye
disease, (2) age between 18 and 64 years, (3) best-corrected visual
acuity of 20/20 or better in each eye, (4) astigmatism of less than
or equal to 3 D, and (5) not using drugs, systemic or ocular, that
may affect vision. Informed consent was collected for participation in the study and, optionally, for being photographed while
using the QS prototype.
Refractions were performed on both eyes of every subject without the use of cycloplegia. Subjects were first measured with a GS
and then with the QS prototype. Lastly, an experienced optometrist
performed binocular SR on the subject, using the GS measurement
as a starting point. We note that starting from the GS measurement
may bias the SR from the optometrist, resulting in a worst-case
comparison of the accuracy of the QS device in matching SR.
The GS measurement was determined from an average of five
measurements of the GS autorefractor set to high accuracy mode
while the subjects looked at a target placed 20¶ away. When subjects
used the QS prototype, they were instructed to hold and look
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Handheld Wavefront AutorefractorVDurr et al.
1143
through the device at an external target placed 20¶ away. Subjects
were asked to manually adjust the position of the device until they
could see a bright red spot and then to look at the target across the
room with both eyes open while a 30-second video was recorded.
Three 30-second videos were recorded for each eye. The first video
was considered a practice run for the subject to align the device. The
analysis presented here was performed on the wavefront images
recorded from the last two videos of each eye.
Analysis of Autorefractor Performance
Refractions obtained from each of the three methods were compared in both the refractive domain (S, C, and A) and the power
vector domain (M, J0, and J45). First, to evaluate the agreement
between the QS and the GS measurements, we perform a BlandAltman analysis on the independent components of the power
vector measurements as well as theqtotal
dioptric strength
of the
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
combined components T DS ¼ M 2 þ J 20 þ J 245 .17 Second,
using SR as the gold standard, we compare the errors in power vector
measurements of the QS and GS autorefractors. In this case, we
define the TDS error as the TDS of the difference between each
component of the power vector measured by objective refraction
and SR. Differences in error distributions were evaluated using a
Wilcoxon signed rank test with a criterion of p less than 0.05 for
statistical significance. Only the results from the right eye for each
subject were considered in the statistical test to avoid the influence
of isometropia on the independence of the samples. Third, we
compare the agreement between the refractions provided by both
autorefractors and SR within thresholds of 0.25 and 0.50 D of
optical power and 10 and 20 degrees of cylindrical orientation.
The axis agreement is calculated by only evaluating eyes that have
a nonzero cylinder measurement by SR (|C | Q 0.25 D).
RESULTS
We tested the accuracy and variability of the QS calibration by
applying the calibration to a test set of three new measurements of
each trial lens used for the calibration. Uncalibrated measurements
tended to slightly underestimate the magnitude of power of the
trial lens for each power vector component except in eyes with
greater than 8 D of hyperopia (Fig. 2). Because the model eyes we
tested have a relatively constant power with smaller apertures, we
were able to measure the correct power of tested trial lenses to over
10 D of hyperopia. The calibrated measurements demonstrated
excellent agreement with the power of the tested trial lenses. In
the linear regime of the spherical equivalent power measurements
(j1 to +1 D), the calibration has an average offset of 0.32 D.
Forty-three subjects were enrolled in the study, but two were
determined to be outside of the M range for inclusion during the
procedure (both were more than j6 D M myopic). Both subjects
were removed from the analysis. The remaining 41 patients (32
were female), ranged in age from 21 to 62 years (26.4 T 9.7 years).
There were no observed adverse events in this study and we were
able to obtain refractions for all recruited subjects using each of the
three refraction methods.
Comparing the power vectors measured from the QS and the
GS with a Bland-Altman analysis, we find average differences of
j0.19, j0.20, 0.09, and 0.06 D for M, J0, J45, and TDS,
FIGURE 2.
QuickSee calibration and test measurements of trial lens combinations that
independently vary the spherical equivalent power, M (A), the vertical
Jackson cross-cylinder power, J0 (B), and the oblique Jackson cross-cylinder
power, J45 (C). The uncalibrated measurements indicate the output of the QS
algorithm using an analytical calculation. These measurements are interpolated piecewise and used to calibrate QS measurements. Calibrated
points show the mean and SD of a set of three separate measurements of
each trial lens combination.
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
1144 Handheld Wavefront AutorefractorVDurr et al.
respectively, with the QS providing lower mean values of M and J0
and higher average values of J45 and TDS than the GS (Fig. 3).
The 95% confidence limits of agreement (1.96 R) between the QS
and GS are T0.85, T0.32, T0.25, and T0.90 D for M, J0, J45, and
TDS, respectively, although the subjects only spanned a small
range of J45 values (j0.38 to 0.26 D when measured by the GS).
Power vector measurements are strongly correlated between the
objective refraction and SR techniques (Fig. 4). The GS and SR
measurements of M were correlated with a Pearson linear correlation coefficient of r = 0.97 (p = 2.4 10j51). The QS and SR
measurements of M were correlated with a Pearson linear correlation coefficient of r = 0.97 (p = 6.4 10j53). The average
difference between M measured by the objective and subjective
technique was j0.08 D for the GS and 0.13 D for the QS. There
are no significant systematic measurement differences associated with the magnitude of the refractive error. However, one
young hyperopic subject was an outlier for the spherical equivalent measurements in both eyes with both objective techniques
compared with the SR. This subject reported to have known
difficulties when obtaining her refractive error as it fluctuates
significantly because of overaccommodation problems. She is a
corrected moderate hyperope, and the SR matched her current
eyeglass correction. Therefore, these values are used to compare
with the objective measurements, both of which showed to be
much less hyperopic.
The mean and SD of the absolute value of the errors for each
autorefractor versus SR are similar (Table 1). Because the error
variables are not normally distributed, we tested for statistical
significance of the differences with a Wilcoxon signed rank test.
There were no statistically significant differences (p 9 0.05 in every
case) in the error distribution of the power vector components
measured with the QS device compared with the GS device, indicating that they have similar accuracy in the power vector domain.
To investigate systematic bias in the astigmatism measured by
each autorefractor compared with SR, we constructed doubleangle plots of the cross-cylinder difference vectors. Fig. 5 shows
that both the QS and the GS autorefractor had asymmetrical
spreads, with the QS exhibiting a center of distribution of 0.12 D
at 161 degrees and the GS exhibiting a center of distribution of
0.10 D at 84 degrees. We also investigate the error rates of each
FIGURE 3.
Bland-Altman plots show the agreement between the QS and GS autorefractors when measuring M (A), J0 (B), J45 (C), and TDS (D). Solid lines represent the
bias between autorefractors (mean of the measurement difference) and dashed lines are placed at the T95% confidence interval for the difference. The two
autorefractors agree within (mean difference T 95% confidence interval) j0.20 T 0.84 D, j0.15 T 0.31 D, 0.06 T 0.25 D, and 0.08 T 0.89 D, for M, J0, J45,
and TDS, respectively.
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Handheld Wavefront AutorefractorVDurr et al.
1145
performance of the QS and GS is more similar, as both devices have
a similar fraction of outliers that deviated from the SR. The axis
measured by the QS and GS were within 10 degrees of SR for 52 and
64% of the eyes, respectively. The poorer performance of the QS for
the axis is likely attributed to variations in how the subjects held the
QS prototypeVthe axis measurement is provided relative to the
orientation of the device and most subjects did not hold the device
consistently vertically.
DISCUSSION
This study evaluates the accuracy of a potentially low-cost,
open-view autorefractor compared with a commonly used commercial autorefractor. One challenge in evaluating the clinical
accuracy of any autorefractor is that although SR is the most relevant gold standard, it is a relatively imprecise technique that can
be confounded by variations in subject feedback, as well as by
uncertainties in the working distance, vertex correction distance,
depth of field, and trial lens power.18 Nevertheless, previous studies
have shown that SR is repeatable within 0.25 D about 80% of the
time,18,19 and introducing as small as a 0.25-D refractive error is
known to cause some discomfort to subjects.20 Therefore, 0.25 D
is a relevant benchmark level of accuracy to attempt to achieve
in autorefraction.
The QS and GS autorefractors tested in this study measured
values of M that agreed within 0.25 D of the value measured from
SR in 71% and 63% of the measured eyes, respectively. This result
is within the range found in other recent studies of commercial
autorefractors, which have shown an agreement of M within
0.25 D of SR in 44%,8,21 51%,8 and 57%22 of the measured eyes.
There are several emerging low-cost and portable technologies currently being pursued to improve refraction in low-resource settings.2
Adjustable lenses can be used for self-refraction and can be locked
to provide the prescription eyeglasses after refraction. However, at
present, they either only correct for defocus errors or, in one case,
require the subject to go through a challenging three-parameter
optimization process to tune S, C, and A for each eye.2,5,6 Another
approach is to retrofit a mobile phone to act as an adjustable
Scheiner disk.7 This strategy has the advantage of being low cost
if a mobile phone with a high-quality screen is already available.
TABLE 1.
Power vector measurement errors of the GS and QS autorefractors
Power vector
M
FIGURE 4.
The M (A), J0 (B), and J45 (C) measured with both the GS and QS
autorefractors correlate well with the SR measurements. The two hyperopic
outliers in the M measurement come from two eyes of the same subject.
autorefractor in comparison to the SR results in the refractive domain. The QS provides spherical and cylindrical powers that more
frequently agreed with the measurements from SR (Table 2). The
QS agrees within 0.25 D of the SR S and C for 74 and 87% of
the eyes, respectively, whereas the GS agrees within 0.25 D of the
SR S and C for 49 and 82% of the eyes. For larger thresholds, the
J0
J45
TDS
Objective test
Mean T SD of absolute
error of right eyes, D
GS
QS
GS
QS
GS
QS
GS
QS
0.40 T 0.46
0.41 T 0.53
0.14 T 0.10
0.16 T 0.11
0.09 T 0.08
0.10 T 0.06
0.47 T 0.44
0.49 T 0.51
p
0.55
0.41
0.18
0.55
Errors are calculated using SR as the gold standard. We found
that there is no statistically significant difference (a p value of 90.05
for each power vector component) between the accuracies of the
two autorefractors in measuring power vectors.
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
1146 Handheld Wavefront AutorefractorVDurr et al.
autorefractor can provide a prescription to about 95% of adults,
depending on the population.24,25 To cover a larger portion of the
population, the range must be expanded to work for patients with
more severe myopia. Additionally, the usability and speed of the
QS could be improved to make alignment easier for the patient
and to reduce the measurement time from 60 seconds (two 30second videos) closer to the approximately 10 seconds that the commercial autorefractors take. These usability and speed constraints are
related to each other in that as the prototype is made more intuitive
and robust to alignment errors, the time required to record a sufficient number of high-quality wavefront images will be reduced.
Improvements in usability and speed will also facilitate the testing
of this device in a more diverse population, including children and
elderly patients.
The accuracy of the QS could also be improved. We found an
average spherical equivalent power error of 0.12 D, which is likely
attributed to chromatic defocus. After testing a larger number of
eyes, this offset can be incorporated into the calibration, and this
systematic error can be reduced. The axis measurement from the
QS was likely limited by the handheld use of the device. Subjects
were instructed to hold the device with the handle vertically but
would often tilt the prototype relative to their face. This limitation
may be overcome by incorporating a contralateral viewing channel
for the opposite eye (a binocular port) so that the axis of the device
is fixed relative to the axis between the two eyes of each subject.
There is also the potential to increase the accuracy of the prototype
by incorporating more sophisticated algorithms that analyze the
higher-order aberrations measured by the device.13,26
To make further progress toward a low-cost autorefractor, several
additional steps are necessary to replace the remaining high-cost
components of the prototype evaluated here: (1) a low-cost microcontroller and display should be incorporated into the device
to replace the laptop currently needed for operation, (2) the 10- 10-mm lenslet array currently used overfills the image sensor and
should be replaced with a smaller clear-aperture lenslet array, and (3)
TABLE 2.
Agreement between refractive parameters measured by
objective refraction and SR
Refractive
parameter
S
C
FIGURE 5.
Double-angle astigmatic distribution of the cross-cylinder difference vectors
between the QS and SR (A) and the GS and SR (B). The centers of the
distributions are located at 0.12 D at 161 degrees and 0.10 D at 84 degrees
from the origin for the QS and GS measurements, respectively. All measurements fell within a 0.50-D-radius circle except for two eyes measured
with the GS autorefractor.
However, in a recent study on 34 eyes, the M measured with the
mobile phone refractometer was found to agree within 0.25 D of the
SR in only 15% of the measured eyes.23
There are several areas in which the QS could be improved. The
range of the current prototype is limited to subjects with an M of
greater than or equal to j6.0 D of myopia. With this range, an
A
M
Objective
test
GS
QS
GS
QS
GS
QS
GS
QS
Agreement with SR
e0.25 D: 49%
e0.25 D: 74%
e0.25 D: 82%
e0.25 D: 87%
e10 degrees: 64%
e10 degrees: 52%
e0.25 D: 63%
e0.25 D: 71%
e0.5 D: 80%
e0.5 D: 82%
e0.5 D: 98%
e0.5 D: 95%
e20 degrees: 85%
e20 degrees: 82%
e0.5 D: 89%
e0.5 D: 82%
The frequency of agreement was calculated for absolute difference thresholds of 0.25 and 0.5 D for power measurements and
10 and 20 degrees for axis measurements. The axis accuracy is
reported from only the set of eyes that had a nonzero cylinder
power (|C| Q 0.25 D) measured by SR. The GS and QS autorefractors measured a spherical equivalent power that was within 0.25 D
of the SR measurement for 63% and 71% of the measured eyes,
respectively.
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.
Handheld Wavefront AutorefractorVDurr et al.
a custom, monolithic housing should be designed to house and orient
the prototype parts.
CONCLUSIONS
This study introduces a novel wavefront-sensing autorefractor
and demonstrates that it is as accurate as a high-end commercial
autorefractor in predicting the power vectors measured by SR in
a population of adults with a moderate range of refractive errors
(j6 D of myopia to +4 D of hyperopia). Given that this autorefractor was implemented in a compact prototype using low-cost
components and no moving parts, it is an encouraging step toward
the development of an autorefractor that is portable, affordable,
and robust.
ACKNOWLEDGMENTS
This work has been financially supported by the Comunidad de Madrid
through the Madrid-MIT M+Visión Consortium. We thank the M+Visión
IDEA3 faculty panel for their assistance in developing this project. We also
thank the subjects who participated in the clinical testing of the QuickSee.
NJD, SRD, DL, CD, SM, and EL are inventors on patents relating
to the autorefractor used in this study and have a financial interest in
PlenOptika, Inc. Since the completion of the study NJD, SRD, DL, and
EL have become part-time employees of PlenOptika.
Received January 21, 2105; accepted July 10, 2015.
REFERENCES
1. Resnikoff S, Pascolini D, Mariotti SP, Pokharel GP. Global magnitude of visual impairment caused by uncorrected refractive errors in
2004. Bull World Health Organ 2008;86:63Y70.
2. Durr NJ, Dave SR, Lage E, Marcos S, Thorn F, Lim D. From unseen
to seen: tackling the global burden of uncorrected refractive errors.
Annu Rev Biomed Eng 2014;16:131Y53.
3. Farook M, Venkatramani J, Gazzard G, Cheng A, Tan D, Saw SM.
Comparisons of the handheld autorefractor, table-mounted autorefractor,
and subjective refraction in Singapore adults. Optom Vis Sci 2005;82:
1066Y70.
4. Choong YF, Chen AH, Goh PP. A comparison of autorefraction and
subjective refraction with and without cycloplegia in primary school
children. Am J Ophthalmol 2006;142:68Y74.
5. Douali MG, Silver JD. Self-optimised vision correction with adaptive
spectacle lenses in developing countries. Ophthalmic Physiol Opt
2004;24:234Y41.
6. Barbero S, Rubinstein J. Power-adjustable sphero-cylindrical refractor comprising two lenses. Opt Eng 2013;52:63002.
7. Pamplona VF, Mohan A, Oliveira MM, Raskar R. NETRA: interactive display for estimating refractive errors and focal range. ACM
Trans Graphic 2010;29:77Y84.
8. Davies LN, Mallen EA, Wolffsohn JS, Gilmartin B. Clinical evaluation of the Shin-Nippon NVision-K 5001/Grand Seiko WR-5100K
autorefractor. Optom Vis Sci 2003;80:320Y4.
9. Cleary G, Spalton DJ, Patel PM, Lin PF, Marshall J. Diagnostic accuracy and variability of autorefraction by the Tracey Visual Function
Analyzer and the Shin-Nippon NVision-K 5001 in relation to subjective refraction. Ophthalmic Physiol Opt 2009;29:173Y81.
1147
10. Delori FC, Webb RH, Sliney DH; American National Standards
Institute. Maximum permissible exposures for ocular safety (ANSI
2000), with emphasis on ophthalmic devices. J Opt Soc Am A 2007;
24:1250Y65.
11. McAlinden C, McCartney M, Moore J. Mathematics of Zernike
polynomials: a review. Clin Experiment Ophthalmol 2011;39:
820Y7.
12. Thibos LN, Wheeler W, Horner D. Power vectors: an application of
Fourier analysis to the description and statistical analysis of refractive
error. Optom Vis Sci 1997;74:367Y75.
13. Thibos LN, Hong X, Bradley A, Applegate RA. Accuracy and precision of objective refraction from wavefront aberrations. J Vis 2004;
4:329Y51.
14. Applegate RA, Ballentine C, Gross H, Sarver EJ, Sarver CA. Visual
acuity as a function of Zernike mode and level of root mean square
error. Optom Vis Sci 2003;80:97Y105.
15. Llorente L, Diaz-Santana L, Lara-Saucedo D, Marcos S. Aberrations
of the human eye in visible and near infrared illumination. Optom
Vis Sci 2003;80:26Y35.
16. Campbell CE. The range of local wavefront curvatures measurable
with Shack-Hartmann wavefront sensors. Clin Exp Optom 2009;
92:187Y93.
17. Raasch TW, Schechtman KB, Davis LJ, Zadnik K; CLEK Study
Group. Collaborative Longitudinal Evaluation of Keratoconus Study.
Repeatability of subjective refraction in myopic and keratoconic subjects: results of vector analysis. Ophthalmic Physiol Opt 2001;21:
376Y83.
18. Smith G. Refraction and visual acuity measurements: what are their
measurement uncertainties? Clin Exp Optom 2006;89:66Y72.
19. Goss DA, Grosvenor T. Reliability of refractionVa literature review.
J Am Optom Assoc 1996;67:619Y30.
20. Miller AD, Kris MJ, Griffiths AC. Effect of small focal errors on
vision. Optom Vis Sci 1997;74:521Y6.
21. Mallen EA, Wolffsohn JS, Gilmartin B, Tsujimura S. Clinical evaluation of the Shin-Nippon SRW-5000 autorefractor in adults. Ophthalmic Physiol Opt 2001;21:101Y7.
22. Sheppard AL, Davies LN. Clinical evaluation of the Grand Seiko
Auto Ref/Keratometer WAM-5500. Ophthalmic Physiol Opt 2010;
30:143Y51.
23. Bastawrous A, Leak C, Howard F, Kumar V. Validation of Near
Eye Tool for Refractive Assessment (NETRA): pilot study. J Mob
Technol Med 2012;1:6Y16.
24. Dandona R, Dandona L, Naduvilath TJ, Srinivas M, McCarty CA,
Rao GN. Refractive errors in an urban population in Southern India:
the Andhra Pradesh Eye Disease Study. Invest Ophthalmol Vis Sci
1999;40:2810Y8.
25. Vitale S, Ellwein L, Cotch MF, Ferris FL 3rd, Sperduto R. Prevalence
of refractive error in the United States, 1999-2004. Arch Ophthalmol
2008;126:1111Y9.
26. Cheng X, Bradley A, Thibos LN. Predicting subjective judgment of
best focus with objective image quality metrics. J Vis 2004;4:310Y21.
Nicholas J. Durr
Johns Hopkins University
Department of Biomedical Engineering
3400 N Charles St, Clark 208D
Baltimore, MD 21218
e-mail: [email protected]
Optometry and Vision Science, Vol. 92, No. 12, December 2015
Copyright © American Academy of Optometry. Unauthorized reproduction of this article is prohibited.